HomeBusinessStudio
Proof-carrying decisions

AI that shows its work — and can be checked.

GraphOS compiles a domain model from your intent and your data, then answers with a decision Thread: evidence on every edge, contracts that say what is missing, and a machine-checkable proof a third party can verify without trusting you — or us.

7/7proofs valid (DPF)
2/2replays reproduce
0.04sto first Thread
5/5benchmark cases

Three verbs, no magic

Ingest evidence, compile the model, solve the decision. Everything in between is inspectable.

Evidence-first graph

Every edge carries a belief id, a source tier and a locator. Unsourced “facts” are downgraded to hypotheses instead of being asserted.

Compiled domain model

Intent plus sample records become an Ontology, rules, evidence contracts and a validation suite — versioned, diffable and approved by a human.

Proof-carrying decisions

Each Thread ships with a DPF document. An independent, stdlib-only checker verifies pins, edge identity, evidence and findings.

What “checked” means here

1 · PinModel hash, snapshot id and ruleset version are frozen with the decision.
2 · CiteEvery accepted fact points to a source and a locator; gaps become findings.
3 · SolveDeterministic or CP-SAT, with a witness for the chosen path or plan.
4 · VerifyRun python -m graphos.proof dpf.json — no GraphOS import, no network.

The evidence pack is the artifact a reviewer reads: status, coverage, findings, evidence, missing evidence, unknowns, and the proof. If a rule cannot run, it is listed as unknown — not hidden.

Pilot pricing

Start with one domain and one reviewer. Success is measured by an external reviewer accepting the evidence pack, not by a demo.

Pilot

€4.9k

4 weeks, one domain, one evidence pack reviewed, weekly working session.

Team

€1.9k/mo

Up to 5 domains, API + Studio, proof export, email support.

Enterprise

Let’s talk

Self-hosted, SSO, audit retention, custom sources and review workflows.